Batch effects in scRNA-seq data

# Batch effects in scRNA-seq data
## Committee meeting
### Almut Lütge
### DMLS - University of Zürich
### 2021-03-02

---

# Batch effects

---
class: inverse, center

# Batch effects

&nbsp;

.pull-left[
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/committee_2/figures/batch1.png" width="80%" height="80%" />
]

.pull-right[
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/committee_2/figures/batch2.png" width="80%" height="80%" />
]

<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/committee_2/figures/batch3.png" width="60%" height="60%" />
---
class: center
# Sinergia: Lung fibroblast from different patients

.pull-left[
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/committee_2/figures/sin_celltype.png" width="100%" height="100%" />
]

--
.pull-right[
<br>
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/committee_2/figures/sin_patient.png" width="87%" height="87%" />
]
---
class: center, middle, inverse

---
class: center, inverse

# Project overview

---
class: center, inverse

# Project overview

---
class: center, inverse

# Project overview

---
class: center, inverse

# Project overview

---
class: center, inverse

# Project overview

---
class: center, inverse

# Project overview

---
class: center, inverse

# Project overview

---
class: inverse, center, middle

# Batch characterization

---

# Batch characterization

&nbsp;

.pull-right[
+ **Variance partitioning**
+ **logFC** distribution and correlation
+ DE genes/overlap
]

&nbsp;
--

### Variance partition

### `$Y_g = \mu + X_{p}\alpha_{pg} + X_{b}\beta_{bg} + X_{p:b} \gamma_{(p:b)g} + \epsilon_g$`

&nbsp;

`$Y_g$`: normalized expression gene g  
`$\mu$`: baseline expression  
`$X_{p}$`, `$X_{b}$`, `$X_{p:b}$`: design matrices for the (random) cell types, batches and interactions effect  
`$\alpha_{pg} \sim N(0, \sigma^2_{pg})$`,..: corresponding random effects  
`$\epsilon_i \sim N(0, \sigma^2_{g})$` corresponding errors

---
class: center

# Percent variance explained by ..:

---
class: center

# Log fold change distributions

---
class: inverse, center, middle

# Metrics

---
class: center, inverse

# Metrics

---
class: center, inverse

# Metrics

---
class: center, inverse

# Metrics

---
class: center

# Cell-specific Mixing Score (cms)

--
.pull-left[
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/CellMixS_final/figures/cms_tsne_X-1.png" width="100%" height="100%" />
]

.pull-right[
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/CellMixS_final/figures/cms_dist_X-1.png" width="100%" height="100%" />
]

---
class: center

# Cell-specific Mixing Score (cms)

--
.pull-left[
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/CellMixS_final/figures/cms_tsne_Y-1.png" width="100%" height="100%" />
]

.pull-right[
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/CellMixS_final/figures/cms_dist_Y-1.png" width="100%" height="100%" />
]

---
class: center
# Sinergia: Lung fibroblast from different patients

.pull-left[
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/committee_2/figures/sin_patient.png" width="80%" height="80%" />
]

.pull-right[
<br>
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/committee_2/figures/sin_cms.png" width="95%" height="95%" />
]

---
class: center
# Sinergia: Lung fibroblast from different patients

.pull-left[
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/committee_2/figures/sin_harmony.png" width="80%" height="80%" />
]

--
.pull-right[
<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/committee_2/figures/sin_harmony_cms.png" width="100%" height="100%" />
]
---
class: inverse, center, middle

# Benchmark

---
class: center

# Task 1: Batch characteristics
<br>
<br>
<br>
<br>
<br>

### *Aim: Test whether metrics reflect batch strength across datasets*

Spearman correlation of metrics with surrogates of batch strength (e.g., PVE-Batch and proportion of DE genes between batches) across datasets

---
class: center

# Batch characteristics
![](data:image/png;base64,#committee2_files/figure-html/unnamed-chunk-27-1.png)

---
class: center

# Task 2: Batch label permutation
<br>
<br>
<br>
<br>
<br>

### *Aim: Negative control and test whether metrics scale with randomness*

Spearman correlation of metrics with the percentage of randomly permuted batch label

---
class: center

# Batch label permutation

# Batch label permutation
<br>

![](data:image/png;base64,#committee2_files/figure-html/unnamed-chunk-29-1.png)

---
class: center

# Task 3: Scaling and detection limits
<br>
<br>
<br>

### *Aim: Test whether metrics scale with (synthetic) batch strength; Estimate lower limit of batch detection*

Spearman correlation of metrics with the batch logFC in simulation series on the same dataset; Minimal batch logFC that is recognized from the metrics as batch effect

---
class:

---
class: center

# Scaling and Sensitivity
![](data:image/png;base64,#committee2_files/figure-html/unnamed-chunk-31-1.png)

---
class: center

# Scaling and Sensitivity

--
<br>

.pull-left[
![](data:image/png;base64,#committee2_files/figure-html/unnamed-chunk-32-1.png)
]

.pull-right[
![](data:image/png;base64,#committee2_files/figure-html/unnamed-chunk-33-1.png)
]

---
class: center

# Task 4: Imbalanced batches
<br>
<br>
<br>

### *Aim: Reaction of metrics to imbalance cell type abundance within the same dataset*

Test sensitivity towards imbalance of cell type abundance

---
class:

---
class: center

# Imbalanced batch effects
<br>
![](data:image/png;base64,#committee2_files/figure-html/unnamed-chunk-35-1.png)

---
class: center

# Imbalanced batch effects

![](data:image/png;base64,#committee2_files/figure-html/unnamed-chunk-36-1.png)

---
class: center, inverse, middle

# Summary

---
class: center

# Summary

![](data:image/png;base64,#committee2_files/figure-html/unnamed-chunk-37-1.png)
---
class: center, inverse, middle

---
class: center, middle, inverse

# 2. *Omni_batch*: open and continuous benchmarking of single cell batch correction methods

---
class: centre

# State of the art

---
class: centre

# Benchmark A

---
class: centre

# Benchmark B

---
class: centre

# Omni-benchmark

---
class: centre

# Omni-benchmark

---
class: centre

# Omni-benchmark

---
class: centre

# Omni-benchmark

---
class: centre

# Omni-benchmark

---
class: centre

# Omni-benchmark

---

# 3. Teaching and course work

---
class:

# Teaching and course work

<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/committee_2/figures/teaching.png" width="95%" height="95%" />
---
---
class: center, middle, inverse

# Extra slides

---
class:

# Local Inverse Simpson Index (lisi)

- neighborhood diversity

- effective number of batches

- neighbor weighting:
  + euclidean distance --> wisi
  + no weighting --> isi
  + Gaussian kernel based weighting --> lisi

`$$\frac{1}{\sum_{b=1}^{B} p(b)}$$`

---
class: center

# k-nearest neighbour batch effect test (kBet)

<img src="data:image/png;base64,#/home/almut/Dokumente/Phd/presentation/CellMixS_final/figures/kbet.png" width="100%" height="100%" />
(Buttner et. al., 2019) 
---
class: center

# Mixing metric (mm)

---